kernel/trace/bpf_trace.c

   1 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of version 2 of the GNU General Public
   5  * License as published by the Free Software Foundation.
   6  */
   7 #include <linux/kernel.h>
   8 #include <linux/types.h>
   9 #include <linux/slab.h>
  10 #include <linux/bpf.h>
  11 #include <linux/filter.h>
  12 #include <linux/uaccess.h>
  13 #include <linux/ctype.h>
  14 #include "trace.h"
  15
  16 /**
  17  * trace_call_bpf - invoke BPF program
  18  * @prog: BPF program
  19  * @ctx: opaque context pointer
  20  *
  21  * kprobe handlers execute BPF programs via this helper.
  22  * Can be used from static tracepoints in the future.
  23  *
  24  * Return: BPF programs always return an integer which is interpreted by
  25  * kprobe handler as:
  26  * 0 - return from kprobe (event is filtered out)
  27  * 1 - store kprobe event into ring buffer
  28  * Other values are reserved and currently alias to 1
  29  */
  30 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
  31 {
  32         unsigned int ret;
  33
  34         if (in_nmi()) /* not supported yet */
  35                 return 1;
  36
  37         preempt_disable();
  38
  39         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
  40                 /*
  41                  * since some bpf program is already running on this cpu,
  42                  * don't call into another bpf program (same or different)
  43                  * and don't send kprobe event into ring-buffer,
  44                  * so return zero here
  45                  */
  46                 ret = 0;
  47                 goto out;
  48         }
  49
  50         rcu_read_lock();
  51         ret = BPF_PROG_RUN(prog, ctx);
  52         rcu_read_unlock();
  53
  54  out:
  55         __this_cpu_dec(bpf_prog_active);
  56         preempt_enable();
  57
  58         return ret;
  59 }
  60 EXPORT_SYMBOL_GPL(trace_call_bpf);
  61
  62 static u64 bpf_probe_read(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
  63 {
  64         void *dst = (void *) (long) r1;
  65         int size = (int) r2;
  66         void *unsafe_ptr = (void *) (long) r3;
  67
  68         return probe_kernel_read(dst, unsafe_ptr, size);
  69 }
  70
  71 static const struct bpf_func_proto bpf_probe_read_proto = {
  72         .func           = bpf_probe_read,
  73         .gpl_only       = true,
  74         .ret_type       = RET_INTEGER,
  75         .arg1_type      = ARG_PTR_TO_STACK,
  76         .arg2_type      = ARG_CONST_STACK_SIZE,
  77         .arg3_type      = ARG_ANYTHING,
  78 };
  79
  80 /*
  81  * limited trace_printk()
  82  * only %d %u %x %ld %lu %lx %lld %llu %llx %p %s conversion specifiers allowed
  83  */
  84 static u64 bpf_trace_printk(u64 r1, u64 fmt_size, u64 r3, u64 r4, u64 r5)
  85 {
  86         char *fmt = (char *) (long) r1;
  87         bool str_seen = false;
  88         int mod[3] = {};
  89         int fmt_cnt = 0;
  90         u64 unsafe_addr;
  91         char buf[64];
  92         int i;
  93
  94         /*
  95          * bpf_check()->check_func_arg()->check_stack_boundary()
  96          * guarantees that fmt points to bpf program stack,
  97          * fmt_size bytes of it were initialized and fmt_size > 0
  98          */
  99         if (fmt[--fmt_size] != 0)
 100                 return -EINVAL;
 101
 102         /* check format string for allowed specifiers */
 103         for (i = 0; i < fmt_size; i++) {
 104                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
 105                         return -EINVAL;
 106
 107                 if (fmt[i] != '%')
 108                         continue;
 109
 110                 if (fmt_cnt >= 3)
 111                         return -EINVAL;
 112
 113                 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
 114                 i++;
 115                 if (fmt[i] == 'l') {
 116                         mod[fmt_cnt]++;
 117                         i++;
 118                 } else if (fmt[i] == 'p' || fmt[i] == 's') {
 119                         mod[fmt_cnt]++;
 120                         i++;
 121                         if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
 122                                 return -EINVAL;
 123                         fmt_cnt++;
 124                         if (fmt[i - 1] == 's') {
 125                                 if (str_seen)
 126                                         /* allow only one '%s' per fmt string */
 127                                         return -EINVAL;
 128                                 str_seen = true;
 129
 130                                 switch (fmt_cnt) {
 131                                 case 1:
 132                                         unsafe_addr = r3;
 133                                         r3 = (long) buf;
 134                                         break;
 135                                 case 2:
 136                                         unsafe_addr = r4;
 137                                         r4 = (long) buf;
 138                                         break;
 139                                 case 3:
 140                                         unsafe_addr = r5;
 141                                         r5 = (long) buf;
 142                                         break;
 143                                 }
 144                                 buf[0] = 0;
 145                                 strncpy_from_unsafe(buf,
 146                                                     (void *) (long) unsafe_addr,
 147                                                     sizeof(buf));
 148                         }
 149                         continue;
 150                 }
 151
 152                 if (fmt[i] == 'l') {
 153                         mod[fmt_cnt]++;
 154                         i++;
 155                 }
 156
 157                 if (fmt[i] != 'd' && fmt[i] != 'u' && fmt[i] != 'x')
 158                         return -EINVAL;
 159                 fmt_cnt++;
 160         }
 161
 162         return __trace_printk(1/* fake ip will not be printed */, fmt,
 163                               mod[0] == 2 ? r3 : mod[0] == 1 ? (long) r3 : (u32) r3,
 164                               mod[1] == 2 ? r4 : mod[1] == 1 ? (long) r4 : (u32) r4,
 165                               mod[2] == 2 ? r5 : mod[2] == 1 ? (long) r5 : (u32) r5);
 166 }
 167
 168 static const struct bpf_func_proto bpf_trace_printk_proto = {
 169         .func           = bpf_trace_printk,
 170         .gpl_only       = true,
 171         .ret_type       = RET_INTEGER,
 172         .arg1_type      = ARG_PTR_TO_STACK,
 173         .arg2_type      = ARG_CONST_STACK_SIZE,
 174 };
 175
 176 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 177 {
 178         /*
 179          * this program might be calling bpf_trace_printk,
 180          * so allocate per-cpu printk buffers
 181          */
 182         trace_printk_init_buffers();
 183
 184         return &bpf_trace_printk_proto;
 185 }
 186
 187 static u64 bpf_perf_event_read(u64 r1, u64 index, u64 r3, u64 r4, u64 r5)
 188 {
 189         struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
 190         struct bpf_array *array = container_of(map, struct bpf_array, map);
 191         struct perf_event *event;
 192         struct file *file;
 193
 194         if (unlikely(index >= array->map.max_entries))
 195                 return -E2BIG;
 196
 197         file = (struct file *)array->ptrs[index];
 198         if (unlikely(!file))
 199                 return -ENOENT;
 200
 201         event = file->private_data;
 202
 203         /* make sure event is local and doesn't have pmu::count */
 204         if (event->oncpu != smp_processor_id() ||
 205             event->pmu->count)
 206                 return -EINVAL;
 207
 208         /*
 209          * we don't know if the function is run successfully by the
 210          * return value. It can be judged in other places, such as
 211          * eBPF programs.
 212          */
 213         return perf_event_read_local(event);
 214 }
 215
 216 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 217         .func           = bpf_perf_event_read,
 218         .gpl_only       = true,
 219         .ret_type       = RET_INTEGER,
 220         .arg1_type      = ARG_CONST_MAP_PTR,
 221         .arg2_type      = ARG_ANYTHING,
 222 };
 223
 224 static u64 bpf_perf_event_output(u64 r1, u64 r2, u64 index, u64 r4, u64 size)
 225 {
 226         struct pt_regs *regs = (struct pt_regs *) (long) r1;
 227         struct bpf_map *map = (struct bpf_map *) (long) r2;
 228         struct bpf_array *array = container_of(map, struct bpf_array, map);
 229         void *data = (void *) (long) r4;
 230         struct perf_sample_data sample_data;
 231         struct perf_event *event;
 232         struct file *file;
 233         struct perf_raw_record raw = {
 234                 .size = size,
 235                 .data = data,
 236         };
 237
 238         if (unlikely(index >= array->map.max_entries))
 239                 return -E2BIG;
 240
 241         file = (struct file *)array->ptrs[index];
 242         if (unlikely(!file))
 243                 return -ENOENT;
 244
 245         event = file->private_data;
 246
 247         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 248                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 249                 return -EINVAL;
 250
 251         if (unlikely(event->oncpu != smp_processor_id()))
 252                 return -EOPNOTSUPP;
 253
 254         perf_sample_data_init(&sample_data, 0, 0);
 255         sample_data.raw = &raw;
 256         perf_event_output(event, &sample_data, regs);
 257         return 0;
 258 }
 259
 260 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 261         .func           = bpf_perf_event_output,
 262         .gpl_only       = true,
 263         .ret_type       = RET_INTEGER,
 264         .arg1_type      = ARG_PTR_TO_CTX,
 265         .arg2_type      = ARG_CONST_MAP_PTR,
 266         .arg3_type      = ARG_ANYTHING,
 267         .arg4_type      = ARG_PTR_TO_STACK,
 268         .arg5_type      = ARG_CONST_STACK_SIZE,
 269 };
 270
 271 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id)
 272 {
 273         switch (func_id) {
 274         case BPF_FUNC_map_lookup_elem:
 275                 return &bpf_map_lookup_elem_proto;
 276         case BPF_FUNC_map_update_elem:
 277                 return &bpf_map_update_elem_proto;
 278         case BPF_FUNC_map_delete_elem:
 279                 return &bpf_map_delete_elem_proto;
 280         case BPF_FUNC_probe_read:
 281                 return &bpf_probe_read_proto;
 282         case BPF_FUNC_ktime_get_ns:
 283                 return &bpf_ktime_get_ns_proto;
 284         case BPF_FUNC_tail_call:
 285                 return &bpf_tail_call_proto;
 286         case BPF_FUNC_get_current_pid_tgid:
 287                 return &bpf_get_current_pid_tgid_proto;
 288         case BPF_FUNC_get_current_uid_gid:
 289                 return &bpf_get_current_uid_gid_proto;
 290         case BPF_FUNC_get_current_comm:
 291                 return &bpf_get_current_comm_proto;
 292         case BPF_FUNC_trace_printk:
 293                 return bpf_get_trace_printk_proto();
 294         case BPF_FUNC_get_smp_processor_id:
 295                 return &bpf_get_smp_processor_id_proto;
 296         case BPF_FUNC_perf_event_read:
 297                 return &bpf_perf_event_read_proto;
 298         case BPF_FUNC_perf_event_output:
 299                 return &bpf_perf_event_output_proto;
 300         case BPF_FUNC_get_stackid:
 301                 return &bpf_get_stackid_proto;
 302         default:
 303                 return NULL;
 304         }
 305 }
 306
 307 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
 308 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type)
 309 {
 310         /* check bounds */
 311         if (off < 0 || off >= sizeof(struct pt_regs))
 312                 return false;
 313
 314         /* only read is allowed */
 315         if (type != BPF_READ)
 316                 return false;
 317
 318         /* disallow misaligned access */
 319         if (off % size != 0)
 320                 return false;
 321
 322         return true;
 323 }
 324
 325 static const struct bpf_verifier_ops kprobe_prog_ops = {
 326         .get_func_proto  = kprobe_prog_func_proto,
 327         .is_valid_access = kprobe_prog_is_valid_access,
 328 };
 329
 330 static struct bpf_prog_type_list kprobe_tl = {
 331         .ops    = &kprobe_prog_ops,
 332         .type   = BPF_PROG_TYPE_KPROBE,
 333 };
 334
 335 static int __init register_kprobe_prog_ops(void)
 336 {
 337         bpf_register_prog_type(&kprobe_tl);
 338         return 0;
 339 }
 340 late_initcall(register_kprobe_prog_ops);